How to add gap detection to the knowledge base you already have

Why can't a knowledge base find its own gaps?

A knowledge base knows two things well: the articles in it and the searches run against it. Both are supply-side records. The articles show what someone decided to write. The search log shows what agents typed. Neither records what customers actually needed, and that is where a knowledge gap lives.

Most knowledge management systems report gaps from search data: zero-result queries, searches with no click, articles rated unhelpful. Those reports are useful for findability, and they only see a gap if an agent happened to search for it. An agent who answered from memory, asked the person next to them, or skipped the search because the knowledge base "never has it" leaves no trace. We cover that blind spot in detail in failed searches vs knowledge gaps.

Adding AI search or answer summaries to the same content doesn't change this. It makes existing articles quicker to reach, including the stale ones, and it still can't tell you which question has no article at all. The knowledge base isn't the problem. It is simply on the wrong side of the conversation to see what is missing.

What does gap detection add to an existing knowledge base?

Gap detection adds the demand side. It reads every customer conversation, records what each customer wanted in plain language, and checks whether an article answers it. The output is a working list for your knowledge team:

• Missing answers: questions customers ask that no article covers.
• Partial or stale answers: an article exists but doesn't fully answer what people now ask.
• Unlinked answers: the article exists and covers the question, but it isn't being matched to it.
• Impact and evidence: how many conversations each gap touched, how difficult they were, and the conversations themselves, so a reviewer can check the finding.

In Knowledge Quest each gap arrives as a recommendation with one of four actions: link an existing article, update one, create a new one, or dismiss with a reason. Create and update start a draft built from the conversations that exposed the gap, grounded in your authoritative source documents, with the sources recorded against the draft. Nothing is published by the recommendation itself. A person approves it first.

Alongside gap detection sit six quality checks that run over the articles you already have: brand alignment, readability, validity, similarity, contradictions and complexity. They catch the problems inside existing content that gap detection, which starts from the questions, would not. See knowledge base content health for how each one works.

What has to connect, and what stays the same?

Two connections do the work. The first is to the place your conversations happen: a contact centre platform such as Genesys Cloud or Amazon Connect, or a service desk such as Freshservice or Zoho Desk. That is where the customer's questions are recorded. The second is to your knowledge base, so each question can be checked against the articles you have and approved fixes can go back.

A lot stays exactly as it is:

• Agents read answers in the same knowledge base, through the same search and the same desktop.
• Your routing, queues and contact flows are untouched. Gap detection reads conversations after they have happened.
• Your publishing rules hold. Articles still need whatever review and approval your process requires before agents see them.
• Your knowledge base stays the source agents trust. Gap detection keeps it complete; it doesn't compete with it.

This is the reason the approach works for teams who have just finished a knowledge base project. The investment in the platform, the article structure and the agent habits is kept. What changes is that the team now works from evidence of what customers asked.

How does live knowledge sync work?

Live knowledge sync keeps your articles aligned between Knowledge Quest and your knowledge base, in both directions. Knowledge Quest syncs live today with these knowledge bases:

• Genesys Knowledge
• LivePro
• Freshservice
• Zoho Desk

Once a knowledge base is connected, sync does four jobs:

1. Brings your articles in. The initial load of existing articles is done during onboarding, so gap detection checks questions against the content agents actually use.
2. Pushes approved work out. New articles and updates drafted from gaps go to your knowledge base once they have been reviewed. Each Knowledge Quest article links to exactly one article on the other side.
3. Pulls edits back. Changes your team makes directly in the knowledge base come back, so the two copies don't drift and the next gap check uses the current wording.
4. Maps your categories. Taxonomy and categories are mapped between the two sides, so gaps, drafts and articles land in the right place in your structure.

Differences are shown on a sync dashboard, row by row: in sync, new on the other side, modified, missing on the other side, or in conflict because both sides changed. For each row someone chooses to push, pull, resolve or set it aside. An alert-only mode reports differences without changing anything, and it is the sensible way to start while the team learns to trust what the comparison shows.

Two safeguards matter in practice. Only reviewed articles should be pushed, because a pushed draft is visible to agents. And if an article disappears from the knowledge base, the Knowledge Quest copy is kept and flagged for its owner; nothing is deleted on the strength of an external removal.

What if your knowledge tool isn't on the sync list?

Gap detection still works. It works alongside your knowledge tool, moving content through import and export instead of a live connection.

Getting articles in. Upload documents (PDF, Word, HTML, Markdown or plain text, several at a time) and each becomes an article, either kept exactly as written or laid out under one of your knowledge templates. Import a public web page by its URL. For larger migrations of structured content, a bulk JSON import is available. Imported articles are then checked against customer questions like any other.

Getting approved work out. Export articles as Word documents, PDF or HTML, either as one combined document or one file per article. A spreadsheet export lists every article's details (title, status, taxonomy path, owner, dates, template alignment, open findings) without the content, which is useful for planning the next batch of updates. Your team then loads the approved articles into the knowledge tool through its own import or editor.

The trade-off is effort, not capability. Gap detection, drafting and the quality checks behave the same either way. Without live sync, someone has to move content across and keep the two copies in step, so it pays to agree a regular cadence, such as a weekly export of approved changes and a monthly re-import of anything edited directly in the knowledge tool.

What does a rollout look like?

1. Connect the conversation source. Map queues or service desk groups to knowledge areas so each conversation is checked against the right set of articles.
2. Bring the articles in. By sync during onboarding, or by import. Add your authoritative source documents (policies, product sheets, procedures) so drafts and validity checks have something to ground against.
3. Agree the taxonomy. Map your existing categories, or let a taxonomy be proposed from your articles and conversations and adjust it, as described in building a taxonomy from conversations.
4. Let evidence build. Gaps appear within days of conversations flowing. The ranking gets sharper as more conversations arrive.
5. Work the first list. Start with the highest-impact gaps. Check whether an existing article can be linked before drafting a new one.
6. Turn on the push. Start sync in alert-only mode, then push approved articles once the team is comfortable with the comparison.

None of these steps needs the knowledge base to be cleaned up first. Running the quality checks early usually turns up duplicates and contradictions worth fixing in parallel. Our guide to duplicate and contradicting articles covers that work.

Who owns what once it is running?

The knowledge team keeps ownership of the content. Gap detection does the reading and ranking, and drafts the fix; people decide what is right and what gets published. A simple split works for most teams:

• Knowledge lead: works the gap list weekly, chooses link, update, create or dismiss, and records the reason.
• Subject-matter experts: review drafts and quality findings in their area before they are approved.
• Knowledge base administrator: looks after sync, resolves conflicts, and keeps categories mapped.
• Team leaders: feed back where agents still struggle, which helps separate a knowledge gap from a training gap.

Reasons on dismissals matter more than they look. They tell the next reviewer why a pattern doesn't warrant an article, and a dismissed gap can resurface if the pattern returns.

How do you know it's working?

Measure coverage and outcomes rather than article counts. Coverage is the share of customer questions that an article answers. The useful number is its change from one period to the next, alongside how many gaps were closed. Behind those, watch the contact outcomes on topics you fixed: transfers, callbacks and repeat contacts should fall. Views and ratings show an article was opened. They don't show it resolved anything. The knowledge gap analysis guide sets out how to rank gaps by impact, and measuring knowledge article impact shows how to check a fix with a fair before-and-after.

This is how Knowledge Quest is built to work. It doesn't replace your contact centre platform or your knowledge base. It reads the conversations, finds what your knowledge base is missing, ranks it by impact and drafts the fix with evidence, then sends approved work to the knowledge base you already run. You keep your tools; they get better knowledge.